The Automatic Identification Of Event Description Unit

With the development of Chinese natural language processing, studies of the Chinese sentences entered the semantic level from the syntactic level. The syntactic and semantic analysis of Chinese sentences becomes an important task in Chinese natural processing.After years of development, the researches on syntax level and semantic level are relatively mature. However, the syntactic and semantic analysis of complete sentences is still difficulties. There are two reasons. One reason is the limitations of existing technologies; another is the characteristic of the Chinese sentence structure. Therefore, the concept of the Event Description Unit (EDU) has been proposed to solve the difficulties of syntax and semantics by some scholars with proposing or improving of method.. That means a complete sentence according to the event is divided into several simple units. Syntactic and Semantic Analysis of complete sentences is transformed to syntactic and semantic analysis of EDU and analysis of logical relationship between EDUs. This method reduces the difficulty of analysis. So, the Identification of the EDU has become an important basis for sentence analysis.Currently, there are two methods of the EDU automatic identification based on statistical and rule. But, they both have some defects. To improve the performance, they have to add new syntactic or semantic features. In this case, we proposed a new method about combining the statistical and rule, and designed two experimental schemes. Without new syntactic and semantic features, we obtained the result that the accuracy is 79.4% and 78.6% in the test. It proved that the identification method based on combining the both is possible.Finally, the paper made the research of syntactic and semantic analysis based on the EDU. Basing on the results of previous studies and combining with the simple structure of the EDU, we proposed syntactic analysis method with the phrase combination rule and CFN element labeling with syntactic information and valence pattern.